-/* memcpy optimized with AVX512 for KNL hardware.\r
- Copyright (C) 2016 Free Software Foundation, Inc.\r
- This file is part of the GNU C Library.\r
-\r
- The GNU C Library is free software; you can redistribute it and/or\r
- modify it under the terms of the GNU Lesser General Public\r
- License as published by the Free Software Foundation; either\r
- version 2.1 of the License, or (at your option) any later version.\r
-\r
- The GNU C Library is distributed in the hope that it will be useful,\r
- but WITHOUT ANY WARRANTY; without even the implied warranty of\r
- MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU\r
- Lesser General Public License for more details.\r
-\r
- You should have received a copy of the GNU Lesser General Public\r
- License along with the GNU C Library; if not, see\r
- <http://www.gnu.org/licenses/>. */\r
-\r
-#include <sysdep.h>\r
-\r
-#if defined HAVE_AVX512_ASM_SUPPORT && IS_IN (libc) \\r
- && (defined SHARED \\r
- || defined USE_AS_MEMMOVE \\r
- || !defined USE_MULTIARCH)\r
-\r
-#include "asm-syntax.h"\r
-#ifndef MEMCPY\r
-# define MEMCPY __memcpy_avx512_no_vzeroupper\r
-# define MEMCPY_CHK __memcpy_chk_avx512_no_vzeroupper\r
-#endif\r
-\r
- .section .text,"ax",@progbits\r
-#if !defined USE_AS_BCOPY\r
-ENTRY (MEMCPY_CHK)\r
- cmpq %rdx, %rcx\r
- jb HIDDEN_JUMPTARGET (__chk_fail)\r
-END (MEMCPY_CHK)\r
-#endif\r
-\r
-ENTRY (MEMCPY)\r
- mov %rdi, %rax\r
-#ifdef USE_AS_MEMPCPY\r
- add %rdx, %rax\r
-#endif\r
- lea (%rsi, %rdx), %rcx\r
- lea (%rdi, %rdx), %r9\r
- cmp $512, %rdx\r
- ja L(512bytesormore)\r
-\r
-L(check):\r
- cmp $16, %rdx\r
- jbe L(less_16bytes)\r
- cmp $256, %rdx\r
- jb L(less_256bytes)\r
- vmovups (%rsi), %zmm0\r
- vmovups 0x40(%rsi), %zmm1\r
- vmovups 0x80(%rsi), %zmm2\r
- vmovups 0xC0(%rsi), %zmm3\r
- vmovups -0x100(%rcx), %zmm4\r
- vmovups -0xC0(%rcx), %zmm5\r
- vmovups -0x80(%rcx), %zmm6\r
- vmovups -0x40(%rcx), %zmm7\r
- vmovups %zmm0, (%rdi)\r
- vmovups %zmm1, 0x40(%rdi)\r
- vmovups %zmm2, 0x80(%rdi)\r
- vmovups %zmm3, 0xC0(%rdi)\r
- vmovups %zmm4, -0x100(%r9)\r
- vmovups %zmm5, -0xC0(%r9)\r
- vmovups %zmm6, -0x80(%r9)\r
- vmovups %zmm7, -0x40(%r9)\r
- ret\r
-\r
-L(less_256bytes):\r
- cmp $128, %dl\r
- jb L(less_128bytes)\r
- vmovups (%rsi), %zmm0\r
- vmovups 0x40(%rsi), %zmm1\r
- vmovups -0x80(%rcx), %zmm2\r
- vmovups -0x40(%rcx), %zmm3\r
- vmovups %zmm0, (%rdi)\r
- vmovups %zmm1, 0x40(%rdi)\r
- vmovups %zmm2, -0x80(%r9)\r
- vmovups %zmm3, -0x40(%r9)\r
- ret\r
-\r
-L(less_128bytes):\r
- cmp $64, %dl\r
- jb L(less_64bytes)\r
- vmovdqu (%rsi), %ymm0\r
- vmovdqu 0x20(%rsi), %ymm1\r
- vmovdqu -0x40(%rcx), %ymm2\r
- vmovdqu -0x20(%rcx), %ymm3\r
- vmovdqu %ymm0, (%rdi)\r
- vmovdqu %ymm1, 0x20(%rdi)\r
- vmovdqu %ymm2, -0x40(%r9)\r
- vmovdqu %ymm3, -0x20(%r9)\r
- ret\r
-\r
-L(less_64bytes):\r
- cmp $32, %dl\r
- jb L(less_32bytes)\r
- vmovdqu (%rsi), %ymm0\r
- vmovdqu -0x20(%rcx), %ymm1\r
- vmovdqu %ymm0, (%rdi)\r
- vmovdqu %ymm1, -0x20(%r9)\r
- ret\r
-\r
-L(less_32bytes):\r
- vmovdqu (%rsi), %xmm0\r
- vmovdqu -0x10(%rcx), %xmm1\r
- vmovdqu %xmm0, (%rdi)\r
- vmovdqu %xmm1, -0x10(%r9)\r
- ret\r
-\r
-L(less_16bytes):\r
- cmp $8, %dl\r
- jb L(less_8bytes)\r
- movq (%rsi), %rsi\r
- movq -0x8(%rcx), %rcx\r
- movq %rsi, (%rdi)\r
- movq %rcx, -0x8(%r9)\r
- ret\r
-\r
-L(less_8bytes):\r
- cmp $4, %dl\r
- jb L(less_4bytes)\r
- mov (%rsi), %esi\r
- mov -0x4(%rcx), %ecx\r
- mov %esi, (%rdi)\r
- mov %ecx, -0x4(%r9)\r
- ret\r
-\r
-L(less_4bytes):\r
- cmp $2, %dl\r
- jb L(less_2bytes)\r
- mov (%rsi), %si\r
- mov -0x2(%rcx), %cx\r
- mov %si, (%rdi)\r
- mov %cx, -0x2(%r9)\r
- ret\r
-\r
-L(less_2bytes):\r
- cmp $1, %dl\r
- jb L(less_1bytes) \r
- mov (%rsi), %cl\r
- mov %cl, (%rdi)\r
-L(less_1bytes):\r
- ret\r
-\r
-L(512bytesormore):\r
-#ifdef SHARED_CACHE_SIZE_HALF\r
- mov $SHARED_CACHE_SIZE_HALF, %r8\r
-#else\r
- mov __x86_shared_cache_size_half(%rip), %r8\r
-#endif\r
- cmp %r8, %rdx\r
- jae L(preloop_large)\r
- cmp $1024, %rdx\r
- ja L(1024bytesormore)\r
- prefetcht1 (%rsi)\r
- prefetcht1 0x40(%rsi)\r
- prefetcht1 0x80(%rsi)\r
- prefetcht1 0xC0(%rsi)\r
- prefetcht1 0x100(%rsi)\r
- prefetcht1 0x140(%rsi)\r
- prefetcht1 0x180(%rsi)\r
- prefetcht1 0x1C0(%rsi)\r
- prefetcht1 -0x200(%rcx)\r
- prefetcht1 -0x1C0(%rcx)\r
- prefetcht1 -0x180(%rcx)\r
- prefetcht1 -0x140(%rcx)\r
- prefetcht1 -0x100(%rcx)\r
- prefetcht1 -0xC0(%rcx)\r
- prefetcht1 -0x80(%rcx)\r
- prefetcht1 -0x40(%rcx) \r
- vmovups (%rsi), %zmm0\r
- vmovups 0x40(%rsi), %zmm1\r
- vmovups 0x80(%rsi), %zmm2\r
- vmovups 0xC0(%rsi), %zmm3\r
- vmovups 0x100(%rsi), %zmm4\r
- vmovups 0x140(%rsi), %zmm5\r
- vmovups 0x180(%rsi), %zmm6\r
- vmovups 0x1C0(%rsi), %zmm7\r
- vmovups -0x200(%rcx), %zmm8\r
- vmovups -0x1C0(%rcx), %zmm9\r
- vmovups -0x180(%rcx), %zmm10\r
- vmovups -0x140(%rcx), %zmm11\r
- vmovups -0x100(%rcx), %zmm12\r
- vmovups -0xC0(%rcx), %zmm13\r
- vmovups -0x80(%rcx), %zmm14\r
- vmovups -0x40(%rcx), %zmm15\r
- vmovups %zmm0, (%rdi)\r
- vmovups %zmm1, 0x40(%rdi)\r
- vmovups %zmm2, 0x80(%rdi)\r
- vmovups %zmm3, 0xC0(%rdi)\r
- vmovups %zmm4, 0x100(%rdi)\r
- vmovups %zmm5, 0x140(%rdi)\r
- vmovups %zmm6, 0x180(%rdi)\r
- vmovups %zmm7, 0x1C0(%rdi)\r
- vmovups %zmm8, -0x200(%r9)\r
- vmovups %zmm9, -0x1C0(%r9)\r
- vmovups %zmm10, -0x180(%r9)\r
- vmovups %zmm11, -0x140(%r9)\r
- vmovups %zmm12, -0x100(%r9)\r
- vmovups %zmm13, -0xC0(%r9)\r
- vmovups %zmm14, -0x80(%r9)\r
- vmovups %zmm15, -0x40(%r9)\r
- ret\r
-\r
-L(1024bytesormore):\r
- cmp %rsi, %rdi\r
- ja L(1024bytesormore_bkw)\r
- sub $512, %r9\r
- vmovups -0x200(%rcx), %zmm8\r
- vmovups -0x1C0(%rcx), %zmm9\r
- vmovups -0x180(%rcx), %zmm10\r
- vmovups -0x140(%rcx), %zmm11\r
- vmovups -0x100(%rcx), %zmm12\r
- vmovups -0xC0(%rcx), %zmm13\r
- vmovups -0x80(%rcx), %zmm14\r
- vmovups -0x40(%rcx), %zmm15\r
- prefetcht1 (%rsi)\r
- prefetcht1 0x40(%rsi)\r
- prefetcht1 0x80(%rsi)\r
- prefetcht1 0xC0(%rsi)\r
- prefetcht1 0x100(%rsi)\r
- prefetcht1 0x140(%rsi)\r
- prefetcht1 0x180(%rsi)\r
- prefetcht1 0x1C0(%rsi)\r
-\r
-/* Loop with unaligned memory access. */\r
-L(gobble_512bytes_loop):\r
- vmovups (%rsi), %zmm0\r
- vmovups 0x40(%rsi), %zmm1\r
- vmovups 0x80(%rsi), %zmm2\r
- vmovups 0xC0(%rsi), %zmm3\r
- vmovups 0x100(%rsi), %zmm4\r
- vmovups 0x140(%rsi), %zmm5\r
- vmovups 0x180(%rsi), %zmm6\r
- vmovups 0x1C0(%rsi), %zmm7\r
- add $512, %rsi\r
- prefetcht1 (%rsi)\r
- prefetcht1 0x40(%rsi)\r
- prefetcht1 0x80(%rsi)\r
- prefetcht1 0xC0(%rsi)\r
- prefetcht1 0x100(%rsi)\r
- prefetcht1 0x140(%rsi)\r
- prefetcht1 0x180(%rsi)\r
- prefetcht1 0x1C0(%rsi)\r
- vmovups %zmm0, (%rdi)\r
- vmovups %zmm1, 0x40(%rdi)\r
- vmovups %zmm2, 0x80(%rdi)\r
- vmovups %zmm3, 0xC0(%rdi)\r
- vmovups %zmm4, 0x100(%rdi)\r
- vmovups %zmm5, 0x140(%rdi)\r
- vmovups %zmm6, 0x180(%rdi)\r
- vmovups %zmm7, 0x1C0(%rdi)\r
- add $512, %rdi\r
- cmp %r9, %rdi\r
- jb L(gobble_512bytes_loop)\r
- vmovups %zmm8, (%r9)\r
- vmovups %zmm9, 0x40(%r9)\r
- vmovups %zmm10, 0x80(%r9)\r
- vmovups %zmm11, 0xC0(%r9)\r
- vmovups %zmm12, 0x100(%r9)\r
- vmovups %zmm13, 0x140(%r9)\r
- vmovups %zmm14, 0x180(%r9)\r
- vmovups %zmm15, 0x1C0(%r9)\r
- ret\r
-\r
-L(1024bytesormore_bkw):\r
- add $512, %rdi\r
- vmovups 0x1C0(%rsi), %zmm8\r
- vmovups 0x180(%rsi), %zmm9\r
- vmovups 0x140(%rsi), %zmm10\r
- vmovups 0x100(%rsi), %zmm11\r
- vmovups 0xC0(%rsi), %zmm12\r
- vmovups 0x80(%rsi), %zmm13\r
- vmovups 0x40(%rsi), %zmm14\r
- vmovups (%rsi), %zmm15\r
- prefetcht1 -0x40(%rcx)\r
- prefetcht1 -0x80(%rcx)\r
- prefetcht1 -0xC0(%rcx)\r
- prefetcht1 -0x100(%rcx)\r
- prefetcht1 -0x140(%rcx)\r
- prefetcht1 -0x180(%rcx)\r
- prefetcht1 -0x1C0(%rcx)\r
- prefetcht1 -0x200(%rcx)\r
- \r
-/* Backward loop with unaligned memory access. */\r
-L(gobble_512bytes_loop_bkw):\r
- vmovups -0x40(%rcx), %zmm0\r
- vmovups -0x80(%rcx), %zmm1\r
- vmovups -0xC0(%rcx), %zmm2\r
- vmovups -0x100(%rcx), %zmm3\r
- vmovups -0x140(%rcx), %zmm4\r
- vmovups -0x180(%rcx), %zmm5\r
- vmovups -0x1C0(%rcx), %zmm6\r
- vmovups -0x200(%rcx), %zmm7\r
- sub $512, %rcx\r
- prefetcht1 -0x40(%rcx)\r
- prefetcht1 -0x80(%rcx)\r
- prefetcht1 -0xC0(%rcx)\r
- prefetcht1 -0x100(%rcx)\r
- prefetcht1 -0x140(%rcx)\r
- prefetcht1 -0x180(%rcx)\r
- prefetcht1 -0x1C0(%rcx)\r
- prefetcht1 -0x200(%rcx)\r
- vmovups %zmm0, -0x40(%r9) \r
- vmovups %zmm1, -0x80(%r9) \r
- vmovups %zmm2, -0xC0(%r9) \r
- vmovups %zmm3, -0x100(%r9) \r
- vmovups %zmm4, -0x140(%r9) \r
- vmovups %zmm5, -0x180(%r9) \r
- vmovups %zmm6, -0x1C0(%r9)\r
- vmovups %zmm7, -0x200(%r9) \r
- sub $512, %r9\r
- cmp %rdi, %r9\r
- ja L(gobble_512bytes_loop_bkw)\r
- vmovups %zmm8, -0x40(%rdi)\r
- vmovups %zmm9, -0x80(%rdi)\r
- vmovups %zmm10, -0xC0(%rdi)\r
- vmovups %zmm11, -0x100(%rdi)\r
- vmovups %zmm12, -0x140(%rdi)\r
- vmovups %zmm13, -0x180(%rdi)\r
- vmovups %zmm14, -0x1C0(%rdi)\r
- vmovups %zmm15, -0x200(%rdi)\r
- ret\r
-\r
-L(preloop_large):\r
- cmp %rsi, %rdi\r
- ja L(preloop_large_bkw)\r
- vmovups (%rsi), %zmm4\r
- vmovups 0x40(%rsi), %zmm5\r
-\r
-/* Align destination for access with non-temporal stores in the loop. */\r
- mov %rdi, %r8\r
- and $-0x80, %rdi\r
- add $0x80, %rdi\r
- sub %rdi, %r8 \r
- sub %r8, %rsi\r
- add %r8, %rdx\r
- prefetcht1 (%rsi)\r
- prefetcht1 0x40(%rsi)\r
- prefetcht1 0x80(%rsi)\r
- prefetcht1 0xC0(%rsi)\r
-L(gobble_256bytes_nt_loop):\r
- vmovups (%rsi), %zmm0\r
- prefetcht1 0x100(%rsi)\r
- vmovups 0x40(%rsi), %zmm1\r
- prefetcht1 0x140(%rsi)\r
- vmovups 0x80(%rsi), %zmm2\r
- prefetcht1 0x180(%rsi)\r
- vmovups 0xC0(%rsi), %zmm3\r
- prefetcht1 0x1C0(%rsi)\r
- vmovntdq %zmm0, (%rdi)\r
- vmovntdq %zmm1, 0x40(%rdi)\r
- vmovntdq %zmm2, 0x80(%rdi)\r
- vmovntdq %zmm3, 0xC0(%rdi)\r
- sub $256, %rdx\r
- add $256, %rsi\r
- add $256, %rdi\r
- cmp $256, %rdx\r
- ja L(gobble_256bytes_nt_loop)\r
- sfence\r
- vmovups %zmm4, (%rax)\r
- vmovups %zmm5, 0x40(%rax)\r
- jmp L(check)\r
-\r
-L(preloop_large_bkw):\r
- vmovups -0x80(%rcx), %zmm4\r
- vmovups -0x40(%rcx), %zmm5\r
-\r
-/* Align end of destination for access with non-temporal stores. */\r
- mov %r9, %r8\r
- and $-0x80, %r9\r
- sub %r9, %r8\r
- sub %r8, %rcx\r
- sub %r8, %rdx\r
- add %r9, %r8\r
- prefetcht1 -0x100(%rcx)\r
- prefetcht1 -0xC0(%rcx)\r
- prefetcht1 -0x80(%rcx)\r
- prefetcht1 -0x40(%rcx)\r
-L(gobble_256bytes_nt_loop_bkw):\r
- vmovups -0x100(%rcx), %zmm0\r
- prefetcht1 -0x200(%rcx)\r
- vmovups -0xC0(%rcx), %zmm1\r
- prefetcht1 -0x1C0(%rcx)\r
- vmovups -0x80(%rcx), %zmm2\r
- prefetcht1 -0x180(%rcx)\r
- vmovups -0x40(%rcx), %zmm3\r
- prefetcht1 -0x140(%rcx)\r
- vmovntdq %zmm0, -0x100(%r9)\r
- vmovntdq %zmm1, -0xC0(%r9)\r
- vmovntdq %zmm2, -0x80(%r9)\r
- vmovntdq %zmm3, -0x40(%r9)\r
- sub $256, %rdx\r
- sub $256, %rcx\r
- sub $256, %r9\r
- cmp $256, %rdx\r
- ja L(gobble_256bytes_nt_loop_bkw)\r
- sfence\r
- vmovups %zmm4, -0x80(%r8)\r
- vmovups %zmm5, -0x40(%r8)\r
- jmp L(check)\r
-\r
-END (MEMCPY)\r
-\r
-#endif\r
+/* memcpy optimized with AVX512 for KNL hardware.
+ Copyright (C) 2016 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#include <sysdep.h>
+
+#if defined HAVE_AVX512_ASM_SUPPORT && IS_IN (libc) \
+ && (defined SHARED \
+ || defined USE_AS_MEMMOVE \
+ || !defined USE_MULTIARCH)
+
+#include "asm-syntax.h"
+#ifndef MEMCPY
+# define MEMCPY __memcpy_avx512_no_vzeroupper
+# define MEMCPY_CHK __memcpy_chk_avx512_no_vzeroupper
+#endif
+
+ .section .text,"ax",@progbits
+#if !defined USE_AS_BCOPY
+ENTRY (MEMCPY_CHK)
+ cmpq %rdx, %rcx
+ jb HIDDEN_JUMPTARGET (__chk_fail)
+END (MEMCPY_CHK)
+#endif
+
+ENTRY (MEMCPY)
+ mov %rdi, %rax
+#ifdef USE_AS_MEMPCPY
+ add %rdx, %rax
+#endif
+ lea (%rsi, %rdx), %rcx
+ lea (%rdi, %rdx), %r9
+ cmp $512, %rdx
+ ja L(512bytesormore)
+
+L(check):
+ cmp $16, %rdx
+ jbe L(less_16bytes)
+ cmp $256, %rdx
+ jb L(less_256bytes)
+ vmovups (%rsi), %zmm0
+ vmovups 0x40(%rsi), %zmm1
+ vmovups 0x80(%rsi), %zmm2
+ vmovups 0xC0(%rsi), %zmm3
+ vmovups -0x100(%rcx), %zmm4
+ vmovups -0xC0(%rcx), %zmm5
+ vmovups -0x80(%rcx), %zmm6
+ vmovups -0x40(%rcx), %zmm7
+ vmovups %zmm0, (%rdi)
+ vmovups %zmm1, 0x40(%rdi)
+ vmovups %zmm2, 0x80(%rdi)
+ vmovups %zmm3, 0xC0(%rdi)
+ vmovups %zmm4, -0x100(%r9)
+ vmovups %zmm5, -0xC0(%r9)
+ vmovups %zmm6, -0x80(%r9)
+ vmovups %zmm7, -0x40(%r9)
+ ret
+
+L(less_256bytes):
+ cmp $128, %dl
+ jb L(less_128bytes)
+ vmovups (%rsi), %zmm0
+ vmovups 0x40(%rsi), %zmm1
+ vmovups -0x80(%rcx), %zmm2
+ vmovups -0x40(%rcx), %zmm3
+ vmovups %zmm0, (%rdi)
+ vmovups %zmm1, 0x40(%rdi)
+ vmovups %zmm2, -0x80(%r9)
+ vmovups %zmm3, -0x40(%r9)
+ ret
+
+L(less_128bytes):
+ cmp $64, %dl
+ jb L(less_64bytes)
+ vmovdqu (%rsi), %ymm0
+ vmovdqu 0x20(%rsi), %ymm1
+ vmovdqu -0x40(%rcx), %ymm2
+ vmovdqu -0x20(%rcx), %ymm3
+ vmovdqu %ymm0, (%rdi)
+ vmovdqu %ymm1, 0x20(%rdi)
+ vmovdqu %ymm2, -0x40(%r9)
+ vmovdqu %ymm3, -0x20(%r9)
+ ret
+
+L(less_64bytes):
+ cmp $32, %dl
+ jb L(less_32bytes)
+ vmovdqu (%rsi), %ymm0
+ vmovdqu -0x20(%rcx), %ymm1
+ vmovdqu %ymm0, (%rdi)
+ vmovdqu %ymm1, -0x20(%r9)
+ ret
+
+L(less_32bytes):
+ vmovdqu (%rsi), %xmm0
+ vmovdqu -0x10(%rcx), %xmm1
+ vmovdqu %xmm0, (%rdi)
+ vmovdqu %xmm1, -0x10(%r9)
+ ret
+
+L(less_16bytes):
+ cmp $8, %dl
+ jb L(less_8bytes)
+ movq (%rsi), %rsi
+ movq -0x8(%rcx), %rcx
+ movq %rsi, (%rdi)
+ movq %rcx, -0x8(%r9)
+ ret
+
+L(less_8bytes):
+ cmp $4, %dl
+ jb L(less_4bytes)
+ mov (%rsi), %esi
+ mov -0x4(%rcx), %ecx
+ mov %esi, (%rdi)
+ mov %ecx, -0x4(%r9)
+ ret
+
+L(less_4bytes):
+ cmp $2, %dl
+ jb L(less_2bytes)
+ mov (%rsi), %si
+ mov -0x2(%rcx), %cx
+ mov %si, (%rdi)
+ mov %cx, -0x2(%r9)
+ ret
+
+L(less_2bytes):
+ cmp $1, %dl
+ jb L(less_1bytes)
+ mov (%rsi), %cl
+ mov %cl, (%rdi)
+L(less_1bytes):
+ ret
+
+L(512bytesormore):
+#ifdef SHARED_CACHE_SIZE_HALF
+ mov $SHARED_CACHE_SIZE_HALF, %r8
+#else
+ mov __x86_shared_cache_size_half(%rip), %r8
+#endif
+ cmp %r8, %rdx
+ jae L(preloop_large)
+ cmp $1024, %rdx
+ ja L(1024bytesormore)
+ prefetcht1 (%rsi)
+ prefetcht1 0x40(%rsi)
+ prefetcht1 0x80(%rsi)
+ prefetcht1 0xC0(%rsi)
+ prefetcht1 0x100(%rsi)
+ prefetcht1 0x140(%rsi)
+ prefetcht1 0x180(%rsi)
+ prefetcht1 0x1C0(%rsi)
+ prefetcht1 -0x200(%rcx)
+ prefetcht1 -0x1C0(%rcx)
+ prefetcht1 -0x180(%rcx)
+ prefetcht1 -0x140(%rcx)
+ prefetcht1 -0x100(%rcx)
+ prefetcht1 -0xC0(%rcx)
+ prefetcht1 -0x80(%rcx)
+ prefetcht1 -0x40(%rcx)
+ vmovups (%rsi), %zmm0
+ vmovups 0x40(%rsi), %zmm1
+ vmovups 0x80(%rsi), %zmm2
+ vmovups 0xC0(%rsi), %zmm3
+ vmovups 0x100(%rsi), %zmm4
+ vmovups 0x140(%rsi), %zmm5
+ vmovups 0x180(%rsi), %zmm6
+ vmovups 0x1C0(%rsi), %zmm7
+ vmovups -0x200(%rcx), %zmm8
+ vmovups -0x1C0(%rcx), %zmm9
+ vmovups -0x180(%rcx), %zmm10
+ vmovups -0x140(%rcx), %zmm11
+ vmovups -0x100(%rcx), %zmm12
+ vmovups -0xC0(%rcx), %zmm13
+ vmovups -0x80(%rcx), %zmm14
+ vmovups -0x40(%rcx), %zmm15
+ vmovups %zmm0, (%rdi)
+ vmovups %zmm1, 0x40(%rdi)
+ vmovups %zmm2, 0x80(%rdi)
+ vmovups %zmm3, 0xC0(%rdi)
+ vmovups %zmm4, 0x100(%rdi)
+ vmovups %zmm5, 0x140(%rdi)
+ vmovups %zmm6, 0x180(%rdi)
+ vmovups %zmm7, 0x1C0(%rdi)
+ vmovups %zmm8, -0x200(%r9)
+ vmovups %zmm9, -0x1C0(%r9)
+ vmovups %zmm10, -0x180(%r9)
+ vmovups %zmm11, -0x140(%r9)
+ vmovups %zmm12, -0x100(%r9)
+ vmovups %zmm13, -0xC0(%r9)
+ vmovups %zmm14, -0x80(%r9)
+ vmovups %zmm15, -0x40(%r9)
+ ret
+
+L(1024bytesormore):
+ cmp %rsi, %rdi
+ ja L(1024bytesormore_bkw)
+ sub $512, %r9
+ vmovups -0x200(%rcx), %zmm8
+ vmovups -0x1C0(%rcx), %zmm9
+ vmovups -0x180(%rcx), %zmm10
+ vmovups -0x140(%rcx), %zmm11
+ vmovups -0x100(%rcx), %zmm12
+ vmovups -0xC0(%rcx), %zmm13
+ vmovups -0x80(%rcx), %zmm14
+ vmovups -0x40(%rcx), %zmm15
+ prefetcht1 (%rsi)
+ prefetcht1 0x40(%rsi)
+ prefetcht1 0x80(%rsi)
+ prefetcht1 0xC0(%rsi)
+ prefetcht1 0x100(%rsi)
+ prefetcht1 0x140(%rsi)
+ prefetcht1 0x180(%rsi)
+ prefetcht1 0x1C0(%rsi)
+
+/* Loop with unaligned memory access. */
+L(gobble_512bytes_loop):
+ vmovups (%rsi), %zmm0
+ vmovups 0x40(%rsi), %zmm1
+ vmovups 0x80(%rsi), %zmm2
+ vmovups 0xC0(%rsi), %zmm3
+ vmovups 0x100(%rsi), %zmm4
+ vmovups 0x140(%rsi), %zmm5
+ vmovups 0x180(%rsi), %zmm6
+ vmovups 0x1C0(%rsi), %zmm7
+ add $512, %rsi
+ prefetcht1 (%rsi)
+ prefetcht1 0x40(%rsi)
+ prefetcht1 0x80(%rsi)
+ prefetcht1 0xC0(%rsi)
+ prefetcht1 0x100(%rsi)
+ prefetcht1 0x140(%rsi)
+ prefetcht1 0x180(%rsi)
+ prefetcht1 0x1C0(%rsi)
+ vmovups %zmm0, (%rdi)
+ vmovups %zmm1, 0x40(%rdi)
+ vmovups %zmm2, 0x80(%rdi)
+ vmovups %zmm3, 0xC0(%rdi)
+ vmovups %zmm4, 0x100(%rdi)
+ vmovups %zmm5, 0x140(%rdi)
+ vmovups %zmm6, 0x180(%rdi)
+ vmovups %zmm7, 0x1C0(%rdi)
+ add $512, %rdi
+ cmp %r9, %rdi
+ jb L(gobble_512bytes_loop)
+ vmovups %zmm8, (%r9)
+ vmovups %zmm9, 0x40(%r9)
+ vmovups %zmm10, 0x80(%r9)
+ vmovups %zmm11, 0xC0(%r9)
+ vmovups %zmm12, 0x100(%r9)
+ vmovups %zmm13, 0x140(%r9)
+ vmovups %zmm14, 0x180(%r9)
+ vmovups %zmm15, 0x1C0(%r9)
+ ret
+
+L(1024bytesormore_bkw):
+ add $512, %rdi
+ vmovups 0x1C0(%rsi), %zmm8
+ vmovups 0x180(%rsi), %zmm9
+ vmovups 0x140(%rsi), %zmm10
+ vmovups 0x100(%rsi), %zmm11
+ vmovups 0xC0(%rsi), %zmm12
+ vmovups 0x80(%rsi), %zmm13
+ vmovups 0x40(%rsi), %zmm14
+ vmovups (%rsi), %zmm15
+ prefetcht1 -0x40(%rcx)
+ prefetcht1 -0x80(%rcx)
+ prefetcht1 -0xC0(%rcx)
+ prefetcht1 -0x100(%rcx)
+ prefetcht1 -0x140(%rcx)
+ prefetcht1 -0x180(%rcx)
+ prefetcht1 -0x1C0(%rcx)
+ prefetcht1 -0x200(%rcx)
+
+/* Backward loop with unaligned memory access. */
+L(gobble_512bytes_loop_bkw):
+ vmovups -0x40(%rcx), %zmm0
+ vmovups -0x80(%rcx), %zmm1
+ vmovups -0xC0(%rcx), %zmm2
+ vmovups -0x100(%rcx), %zmm3
+ vmovups -0x140(%rcx), %zmm4
+ vmovups -0x180(%rcx), %zmm5
+ vmovups -0x1C0(%rcx), %zmm6
+ vmovups -0x200(%rcx), %zmm7
+ sub $512, %rcx
+ prefetcht1 -0x40(%rcx)
+ prefetcht1 -0x80(%rcx)
+ prefetcht1 -0xC0(%rcx)
+ prefetcht1 -0x100(%rcx)
+ prefetcht1 -0x140(%rcx)
+ prefetcht1 -0x180(%rcx)
+ prefetcht1 -0x1C0(%rcx)
+ prefetcht1 -0x200(%rcx)
+ vmovups %zmm0, -0x40(%r9)
+ vmovups %zmm1, -0x80(%r9)
+ vmovups %zmm2, -0xC0(%r9)
+ vmovups %zmm3, -0x100(%r9)
+ vmovups %zmm4, -0x140(%r9)
+ vmovups %zmm5, -0x180(%r9)
+ vmovups %zmm6, -0x1C0(%r9)
+ vmovups %zmm7, -0x200(%r9)
+ sub $512, %r9
+ cmp %rdi, %r9
+ ja L(gobble_512bytes_loop_bkw)
+ vmovups %zmm8, -0x40(%rdi)
+ vmovups %zmm9, -0x80(%rdi)
+ vmovups %zmm10, -0xC0(%rdi)
+ vmovups %zmm11, -0x100(%rdi)
+ vmovups %zmm12, -0x140(%rdi)
+ vmovups %zmm13, -0x180(%rdi)
+ vmovups %zmm14, -0x1C0(%rdi)
+ vmovups %zmm15, -0x200(%rdi)
+ ret
+
+L(preloop_large):
+ cmp %rsi, %rdi
+ ja L(preloop_large_bkw)
+ vmovups (%rsi), %zmm4
+ vmovups 0x40(%rsi), %zmm5
+
+/* Align destination for access with non-temporal stores in the loop. */
+ mov %rdi, %r8
+ and $-0x80, %rdi
+ add $0x80, %rdi
+ sub %rdi, %r8
+ sub %r8, %rsi
+ add %r8, %rdx
+ prefetcht1 (%rsi)
+ prefetcht1 0x40(%rsi)
+ prefetcht1 0x80(%rsi)
+ prefetcht1 0xC0(%rsi)
+L(gobble_256bytes_nt_loop):
+ prefetcht1 0x200(%rsi)
+ prefetcht1 0x240(%rsi)
+ prefetcht1 0x280(%rsi)
+ prefetcht1 0x2C0(%rsi)
+ prefetchnta 0x300(%rsi)
+ prefetchnta 0x340(%rsi)
+ prefetchnta 0x380(%rsi)
+ prefetchnta 0x3C0(%rsi)
+ vmovups (%rsi), %zmm0
+ vmovups 0x40(%rsi), %zmm1
+ vmovups 0x80(%rsi), %zmm2
+ vmovups 0xC0(%rsi), %zmm3
+ vmovntdq %zmm0, (%rdi)
+ vmovntdq %zmm1, 0x40(%rdi)
+ vmovntdq %zmm2, 0x80(%rdi)
+ vmovntdq %zmm3, 0xC0(%rdi)
+ sub $256, %rdx
+ add $256, %rsi
+ add $256, %rdi
+ cmp $256, %rdx
+ ja L(gobble_256bytes_nt_loop)
+ sfence
+ vmovups %zmm4, (%rax)
+ vmovups %zmm5, 0x40(%rax)
+ jmp L(check)
+
+L(preloop_large_bkw):
+ vmovups -0x80(%rcx), %zmm4
+ vmovups -0x40(%rcx), %zmm5
+
+/* Align end of destination for access with non-temporal stores. */
+ mov %r9, %r8
+ and $-0x80, %r9
+ sub %r9, %r8
+ sub %r8, %rcx
+ sub %r8, %rdx
+ add %r9, %r8
+L(gobble_256bytes_nt_loop_bkw):
+ prefetcht1 -0x300(%rcx)
+ prefetcht1 -0x2C0(%rcx)
+ prefetcht1 -0x280(%rcx)
+ prefetcht1 -0x240(%rcx)
+ prefetchnta -0x400(%rcx)
+ prefetchnta -0x3C0(%rcx)
+ prefetchnta -0x380(%rcx)
+ prefetchnta -0x340(%rcx)
+ vmovups -0x100(%rcx), %zmm0
+ vmovups -0xC0(%rcx), %zmm1
+ vmovups -0x80(%rcx), %zmm2
+ vmovups -0x40(%rcx), %zmm3
+ vmovntdq %zmm0, -0x100(%r9)
+ vmovntdq %zmm1, -0xC0(%r9)
+ vmovntdq %zmm2, -0x80(%r9)
+ vmovntdq %zmm3, -0x40(%r9)
+ sub $256, %rdx
+ sub $256, %rcx
+ sub $256, %r9
+ cmp $256, %rdx
+ ja L(gobble_256bytes_nt_loop_bkw)
+ sfence
+ vmovups %zmm4, -0x80(%r8)
+ vmovups %zmm5, -0x40(%r8)
+ jmp L(check)
+END (MEMCPY)
+#endif
projects / thirdparty / glibc.git / commitdiff
